If we could support a larger file, we can just zero pad the counts (stored binary) to max needed length. Then we get accurate seek to entry # x. If that is not

If we could support a larger file, we can just zero pad the counts (stored binary) to max needed length. Then we get accurate seek to entry # x. If that is not possible, precompute the average count representation length to get an accurate AVG offset per entry. Accurate in aggregate. But I wonder if I wouldn’t use some prefix representation, like a trie, but that’s cheating I think